Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

My computer's Ethernet adapter uses 1,500 byte Ethernet frames. Normally this means that my Local Area Connection's MTU is 1500 bytes.

But my (IPv4) connection to the Internet has some overhead. This means that my actual MTU when communicating to the Internet is 1492. This is causing all the regular problems that exist with mis-configured interface MTU.

The Router Can Fix It

Obviously i could use netsh, and override my network interface with a custom MTU. But i don't want to do that. My router's DHCP server already knows that the MTU is 1492. If asked, the DHCP Server will respond with DHCP option 26:

  • 26 MTU Interface = 1492

So it was driving me nuts why the DHCP server isn't telling my Windows 7 machine the MTU setting as one of the items it returns in the DHCP offer:

  • 54 DHCP Server Identifier = 192.168.1.1
  • 51 IP Address Lease Time = 2 hours
  • 1 Subnet Mask = 255.255.255.0
  • 15 Domain Name = example.local
  • 3 Router = 192.168.1.1
  • 6 Domain Name Server = 8.8.8.8, 8.8.4.4, 74.82.42.42

No MTU Interface! So my machine's Ethernet is stuck at 1500 bytes.

DHCP server's did what it's supposed to do

It took more digging, but i realized that DHCP Server's only give what the client asked for. During my Windows 7 DHCP Discover, Windows 7 enumerates the list of all the DHCP options it would like returned:

  • 1 Subnet Mask
  • 15 Domain Name
  • 3 Router
  • 6 Domain Name Server
  • 44 NetBIOS over TCP/IP Name Server
  • 46 NetBIOS over TCP/IP Node Type
  • 47 NetBIOS over TCP/IP Scope
  • 31 Perform Router Discovery
  • 33 Static Route
  • 121 Classless Static Route
  • 249 Classless Static Route (Microsoft)
  • 43 Vendor-Specific Information

Windows 7 isn't even asking for the MTU!

It must know what it's doing

Windows must have a reason why it's not asking for the MTU; but i don't know what it is. At first i thought it was because it was using Path Maximum MTU Discovery. But PMTU only works for TCP connections, uses the TCP MSS (Maximum Segment Size) option, and requires that the local MSS option be correct.

In my case the TCP MSS setting (which comes from the MTU option) is wrong. It is 1460 bytes

  1500 interface MTU
-   20 bytes IP header = 1480 bytes
-   20 bytes TCP header
= 1460 bytes MSS

which is too high.

Note: I can confirm Windows is using an MSS of 1460 bytes from a SYN packet, when a TCP connection is being established.

If not PathMTU, then what?

What is Windows 7 thinking? What is the intended behavior here? What were the network stack developers at Microsoft thinking should happen?

Now pretend that my Internet MTU is actually 1472 (which it is), and pretend that my Ethernet adapter actually uses 8,192 byte Ethernet frames (which it does). What is a corporation supposed to do? Am i supposed to walk up to every machine in the enterprise and type:

>netsh interface ipv4 set interface interface="Local Area Connection" mtu=1472 store=persistent

That's not reasonable. And even if it were reasonable, it's not what i'm asking.

I'm trying to understand what Windows wants to have happen. i'm trying to understand how it should behave. i'm trying to learn something here. Pretend that my Internet gateway is a 576 byte ATM; how do i instruct Windows 7 machines that the MTU if they want to send to the Internet is 576 bytes?

Bonus Reading

share|improve this question
    
According to experts-exchange.com/OS/Miscellaneous/Q_21711433.html (right down the bottom) someone anecdotally reports that Windows doesn't ask for the MTU (which is what you're seeing). As far as what a corporation would do, they'd probably use Group Policy to run a script that would apply the changes. –  ta.speot.is Oct 7 '13 at 0:44
1  
For what a corporation would do, the router would have its outside MTU set properly after testing maximum mtu allowed with ping (typically 1380-1500). The inside, being pure ethernet would be 1500 and the Windows machines left alone as they communicate to the router on an ethernet network. –  Fiasco Labs Oct 7 '13 at 0:56
    
@FiascoLabs But what happens when the router drops packets destined for the Internet when they cannot fit out the pipe? –  Ian Boyd Oct 7 '13 at 14:07
add comment

1 Answer 1

up vote 3 down vote accepted

Path MTU is not TCP only and doesn't use MSS (though as you have surmised, MSS is derived from MTU). Any hop that receives a packet too big to forward will send an ICMP fragmentation needed packet back to the origin. This is part of layer 3 and is independent of TCP.

The problem with PMTU is that the incoming ICMP packets are often dropped or not forwarded to the origin, and so the origin does not know to reduce the packet size.

As your router knows that the next hop MTU is 1472, it should be sending PMTU packets to your clients. You can check this with wireshark, and if it is not doing this, it needs fixing.

You have a couple of workarounds. One is to set the inside interface of the router's MTU to 1472, which will mean it gets actioned before the router does any processing. The other option is (if you can do it with your router) is to turn on the DF bit for every packet. This is a pretty brutal workaround as any full size packets will end up as two separate packets, one small, one large, with the resultant overhead.

In a corporate setting, all three approaches may be used, but hard setting mtu on the router LAN interface is pretty common.

(I have never come across an instance where DHCP is used to mediate MTU, it wouldn't be the right place in most cases, because the internal MTU and border MTU often different, where you might use jumbo frames internally, but need to reduce to normal size for internet traffic)

share|improve this answer
    
i got my (mis)information about Path MTU Discovery relying on TCP MSS from this MSDN article. i see now that just read it wrong. Consulting RFC1191 - Path MTU Discovery would have been the better choice. Now i'm left to wonder why the local router isn't responding with ICMP Destination Unreachable - Fragmentation needed but DF flag set. pfSense just released version 2.1, and it doesn't send ICMP 3-4 :( –  Ian Boyd Oct 7 '13 at 0:58
    
Is pfSense just dropping icmp unreachables perhaps? This is often set as a "security" measure with some devices, but often causes more trouble than it saves. –  Paul Oct 7 '13 at 1:22
    
i've been preparing screenshots and documentation for that very issue in pfSense. Unfortunately they are pretty testy is someone suggests an issue with their system. Then they'd likely suggest that i shouldn't be relying on ICMP notifications; as a lot of firewalls are misconfigured to block it. And finally that i should simply set the MTU on enterprise's LAN NICs. –  Ian Boyd Oct 7 '13 at 14:07
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.